how to train? about region-based-non-local-network HOT 2 CLOSED

For the training on Something-V1 dataset. I use the script below:
python main.py --dist-url 'tcp://localhost:6666' --dist-backend 'nccl' --multiprocessing-distributed --available_gpus 0,1,2,3 --world-size 1 --rank 0 --dataset something --gd 20 --shift --shift_div=8 --shift_place=blockres --npb --lr 0.02 --wd 1e-3 --dropout 0.8 --num_segments 8 --batch_size 16 --batch_multiplier 4 --use_warmup --warmup_epochs 1 --lr_type cos --epochs 50 --non_local --suffix 1
you should get the same result on the paper. The trained models will release recently.

About how to organize data, the code of my data is acturally based on https://github.com/yjxiong/tsn-pytorch, which can provide you more details.

from region-based-non-local-network.

Thank you so much!

from region-based-non-local-network.